Upload folder using huggingface_hub

README.md

@@ -1,13 +1,27 @@
 ---
-title: Gliner2 5 Multi Extractor
-emoji: ⚡
-colorFrom: yellow
-colorTo: blue
+title: GLiNER2.5 Multi — Zero-Shot Information Extraction
+emoji: 🔍
+colorFrom: green
+colorTo: pink
 sdk: gradio
 sdk_version: 6.25.0
-python_version: '3.12'
 app_file: app.py
-pinned: false
+short_description: Zero-shot entity, classification, relation extraction
+python_version: "3.12"
+startup_duration_timeout: 30m
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+Zero-shot multilingual information extraction with [fastino/gliner2.5-multi-v1](https://huggingface.co/fastino/gliner2.5-multi-v1).
+
+Supports four tasks, all with user-defined labels — no retraining needed:
+
+- **Entity Extraction** — detect named entities (person, organization, product, …)
+- **Text Classification** — single or multi-label zero-shot classification
+- **Relation Extraction** — identify relationships between entities
+- **Structured Data Extraction** — parse unstructured text into typed JSON records
+
+The model is a 287M-parameter boundary-architecture encoder (mDeBERTa-v3-base) supporting 4096-token context and multiple languages.
+
+## License
+
+Apache-2.0

app.py

@@ -0,0 +1,312 @@
+import spaces  # MUST come before torch / any CUDA-touching import
+import torch
+import gradio as gr
+import json
+import re
+
+from gliner2 import AutoExtractor
+
+MODEL_ID = "fastino/gliner2.5-multi-v1"
+
+model = AutoExtractor.from_pretrained(MODEL_ID, map_location="cuda")
+model.eval()
+
+CSS = """
+#col-container { max-width: 1100px; margin: 0 auto; }
+.dark .gradio-container { color: var(--body-text-color); }
+"""
+
+
+def _parse_labels(labels_text):
+    """Parse comma-separated labels into a clean list."""
+    if not labels_text or not labels_text.strip():
+        return []
+    labels = [l.strip() for l in labels_text.split(",") if l.strip()]
+    return labels
+
+
+def _parse_class_schema(schema_text):
+    """Parse classification schema from text like: sentiment: positive, negative, neutral"""
+    result = {}
+    if not schema_text or not schema_text.strip():
+        return result
+    for line in schema_text.strip().split("\n"):
+        if ":" in line:
+            task, labels_str = line.split(":", 1)
+            task = task.strip()
+            labels = [l.strip() for l in labels_str.split(",") if l.strip()]
+            if task and labels:
+                result[task] = labels
+    return result
+
+
+def _format_json(obj):
+    """Pretty-print JSON for display."""
+    return json.dumps(obj, indent=2, ensure_ascii=False, default=str)
+
+
+@spaces.GPU(duration=30)
+def extract_entities(text, labels_text, include_confidence=True, include_spans=True):
+    """Extract named entities from text using zero-shot GLiNER2.5.
+
+    Args:
+        text: The input text to extract entities from.
+        labels_text: Comma-separated entity labels to detect (e.g. "person, organization, location").
+        include_confidence: Whether to include confidence scores in output.
+        include_spans: Whether to include character spans in output.
+    """
+    labels = _parse_labels(labels_text)
+    if not text.strip():
+        return "Please enter some text."
+    if not labels:
+        return "Please enter at least one entity label."
+    result = model.extract_entities(
+        text,
+        labels,
+        include_confidence=include_confidence,
+        include_spans=include_spans,
+    )
+    return _format_json(result)
+
+
+@spaces.GPU(duration=30)
+def classify_text(text, schema_text):
+    """Classify text into categories using zero-shot classification with GLiNER2.5.
+
+    Args:
+        text: The input text to classify.
+        schema_text: Classification schema, one task per line in format 'task: label1, label2, ...'.
+    """
+    schema = _parse_class_schema(schema_text)
+    if not text.strip():
+        return "Please enter some text."
+    if not schema:
+        return "Please enter a classification schema (e.g. 'sentiment: positive, negative, neutral')."
+    result = model.classify_text(text, schema)
+    return _format_json(result)
+
+
+@spaces.GPU(duration=30)
+def extract_relations(text, labels_text, include_confidence=True, include_spans=True):
+    """Extract relations between entities from text using GLiNER2.5.
+
+    Args:
+        text: The input text to extract relations from.
+        labels_text: Comma-separated relation labels to detect (e.g. "works_for, located_in").
+        include_confidence: Whether to include confidence scores.
+        include_spans: Whether to include character spans.
+    """
+    labels = _parse_labels(labels_text)
+    if not text.strip():
+        return "Please enter some text."
+    if not labels:
+        return "Please enter at least one relation label."
+    result = model.extract_relations(
+        text,
+        labels,
+        include_confidence=include_confidence,
+        include_spans=include_spans,
+    )
+    return _format_json(result)
+
+
+@spaces.GPU(duration=30)
+def extract_structured(text, schema_text):
+    """Extract structured JSON data from text using GLiNER2.5.
+
+    Args:
+        text: The input text to extract structured data from.
+        schema_text: JSON schema description, one field per line in format 'field: type::description'.
+    """
+    if not text.strip():
+        return "Please enter some text."
+    # Parse schema: field_name::type::description (one per line)
+    schema = {}
+    for line in schema_text.strip().split("\n"):
+        line = line.strip()
+        if not line:
+            continue
+        parts = line.split("::", 2)
+        if len(parts) >= 1:
+            field = parts[0].strip()
+            dtype = parts[1].strip() if len(parts) > 1 else "str"
+            desc = parts[2].strip() if len(parts) > 2 else ""
+            entry = f"{dtype}::{desc}" if desc else dtype
+            schema[field] = [entry] if dtype != "list" else [f"list::{desc}" if desc else "list"]
+    if not schema:
+        return "Please enter a schema (e.g. 'name::str::Product name')."
+    result = model.extract_json(text, schema)
+    return _format_json(result)
+
+
+with gr.Blocks(theme=gr.themes.Citrus(), css=CSS, title="GLiNER2.5 Multi — Information Extraction") as demo:
+    gr.Markdown("""
+    # 🔍 GLiNER2.5 Multi — Zero-Shot Information Extraction
+
+    Multilingual, multi-task information extraction with [fastino/gliner2.5-multi-v1](https://huggingface.co/fastino/gliner2.5-multi-v1) (287M params, mDeBERTa-v3 encoder).
+
+    Define your own labels at inference time — no retraining needed. Supports entity recognition, text classification, relation extraction, and structured data extraction across multiple languages.
+    """)
+
+    with gr.Row():
+        with gr.Column():
+            gr.Markdown("## 🏷️ Entity Extraction")
+            gr.Markdown("Extract named entities with custom labels.")
+            ner_text = gr.Textbox(
+                label="Input Text",
+                placeholder="Enter text to analyze…",
+                lines=5,
+                value="Apple CEO Tim Cook announced iPhone 15 in Cupertino yesterday. The event was held at Apple Park.",
+            )
+            ner_labels = gr.Textbox(
+                label="Entity Labels (comma-separated)",
+                value="company, person, product, location",
+                placeholder="person, organization, location…",
+            )
+            with gr.Accordion("Options", open=False):
+                ner_conf = gr.Checkbox(label="Include confidence scores", value=True)
+                ner_spans = gr.Checkbox(label="Include character spans", value=True)
+            ner_btn = gr.Button("Extract Entities", variant="primary")
+            ner_output = gr.Code(label="Result (JSON)", language="json", lines=15)
+
+            gr.Examples(
+                examples=[
+                    ["Apple CEO Tim Cook announced iPhone 15 in Cupertino yesterday. The event was held at Apple Park.", "company, person, product, location"],
+                    ["Barcelona defeated Real Madrid 3-1 at Camp Nou. Lewandowski scored twice for Barça.", "team, player, city, stadium"],
+                    ["Marie Curie was born in Warsaw and later moved to Paris to work at the Sorbonne.", "person, city, country, organization"],
+                ],
+                inputs=[ner_text, ner_labels],
+                outputs=ner_output,
+                fn=extract_entities,
+                cache_examples=True,
+                cache_mode="lazy",
+            )
+
+    gr.Markdown("---")
+
+    with gr.Row():
+        with gr.Column():
+            gr.Markdown("## 📋 Text Classification")
+            gr.Markdown("Classify text into custom categories (zero-shot).")
+            cls_text = gr.Textbox(
+                label="Input Text",
+                placeholder="Enter text to classify…",
+                lines=3,
+                value="This laptop has amazing performance but terrible battery life!",
+            )
+            cls_schema = gr.Textbox(
+                label="Classification Schema (one task per line: task: label1, label2, …)",
+                value="sentiment: positive, negative, neutral",
+                lines=3,
+            )
+            cls_btn = gr.Button("Classify Text", variant="primary")
+            cls_output = gr.Code(label="Result (JSON)", language="json", lines=8)
+
+            gr.Examples(
+                examples=[
+                    ["This laptop has amazing performance but terrible battery life!", "sentiment: positive, negative, neutral"],
+                    ["Breaking: Tech giant acquires AI startup for $2B in landmark deal.", "topic: technology, business, politics, sports"],
+                    ["Le film était captivant du début à la fin, avec des acteurs brillants.", "sentiment: positif, négatif, neutre"],
+                ],
+                inputs=[cls_text, cls_schema],
+                outputs=cls_output,
+                fn=classify_text,
+                cache_examples=True,
+                cache_mode="lazy",
+            )
+
+    gr.Markdown("---")
+
+    with gr.Row():
+        with gr.Column():
+            gr.Markdown("## 🔗 Relation Extraction")
+            gr.Markdown("Detect relationships between entities in text.")
+            rel_text = gr.Textbox(
+                label="Input Text",
+                placeholder="Enter text to analyze…",
+                lines=4,
+                value="Alice works for Acme Corp in Paris. Bob joined Acme last year and lives in London.",
+            )
+            rel_labels = gr.Textbox(
+                label="Relation Labels (comma-separated)",
+                value="works_for, located_in",
+                placeholder="works_for, located_in, founded_by…",
+            )
+            with gr.Accordion("Options", open=False):
+                rel_conf = gr.Checkbox(label="Include confidence scores", value=True)
+                rel_spans = gr.Checkbox(label="Include character spans", value=True)
+            rel_btn = gr.Button("Extract Relations", variant="primary")
+            rel_output = gr.Code(label="Result (JSON)", language="json", lines=15)
+
+            gr.Examples(
+                examples=[
+                    ["Alice works for Acme Corp in Paris. Bob joined Acme last year and lives in London.", "works_for, located_in"],
+                    ["Google was founded by Larry Page and Sergey Brin in Mountain View.", "founded_by, located_in"],
+                    ["John Smith married Jane Doe in 2015 in New York City.", "married_to, located_in"],
+                ],
+                inputs=[rel_text, rel_labels],
+                outputs=rel_output,
+                fn=extract_relations,
+                cache_examples=True,
+                cache_mode="lazy",
+            )
+
+    gr.Markdown("---")
+
+    with gr.Row():
+        with gr.Column():
+            gr.Markdown("## 📦 Structured Data Extraction")
+            gr.Markdown("Parse text into structured JSON records with typed fields.")
+            json_text = gr.Textbox(
+                label="Input Text",
+                placeholder="Enter text to extract structured data from…",
+                lines=4,
+                value="iPhone 15 Pro Max with 256GB storage, A17 Pro chip, priced at $1199. Available in titanium and black colors.",
+            )
+            json_schema = gr.Textbox(
+                label="Schema (one field per line: field::type::description)",
+                value="name::str::Full product name and model\nstorage::str::Storage capacity\nprocessor::str::Chip or processor\nprice::str::Product price with currency\ncolors::list::Available color options",
+                lines=5,
+            )
+            json_btn = gr.Button("Extract Structured Data", variant="primary")
+            json_output = gr.Code(label="Result (JSON)", language="json", lines=12)
+
+            gr.Examples(
+                examples=[
+                    ["iPhone 15 Pro Max with 256GB storage, A17 Pro chip, priced at $1199. Available in titanium and black colors.", "name::str::Full product name and model\nstorage::str::Storage capacity\nprocessor::str::Chip or processor\nprice::str::Product price with currency\ncolors::list::Available color options"],
+                    ["Alice bought apples for $3.50 at Whole Foods. Bob purchased oranges for $2.00 at Trader Joe's.", "buyer::str::Name of buyer\nitem::str::Item purchased\nprice::str::Price paid\nstore::str::Store name"],
+                ],
+                inputs=[json_text, json_schema],
+                outputs=json_output,
+                fn=extract_structured,
+                cache_examples=True,
+                cache_mode="lazy",
+            )
+
+    # Wire buttons
+    ner_btn.click(
+        extract_entities,
+        inputs=[ner_text, ner_labels, ner_conf, ner_spans],
+        outputs=ner_output,
+        api_name="extract_entities",
+    )
+    cls_btn.click(
+        classify_text,
+        inputs=[cls_text, cls_schema],
+        outputs=cls_output,
+        api_name="classify_text",
+    )
+    rel_btn.click(
+        extract_relations,
+        inputs=[rel_text, rel_labels, rel_conf, rel_spans],
+        outputs=rel_output,
+        api_name="extract_relations",
+    )
+    json_btn.click(
+        extract_structured,
+        inputs=[json_text, json_schema],
+        outputs=json_output,
+        api_name="extract_structured",
+    )
+
+demo.launch(mcp_server=True)

requirements.txt

@@ -0,0 +1 @@
+gliner2[local]